Automatic driving intelligent acceleration and deceleration lane changing decision-making method

Abstract

The invention discloses an automatic driving intelligent acceleration and deceleration lane changing decision-making method. The method comprises three processes of lane changing intention generation, acceleration and deceleration lane changing strategy, and extreme scene identification and processing. The method is different from a traditional single method and belongs to a mixed method, the advantages of simplicity and good interpretability of a rule-based method are integrated, and meanwhile, an efficient optimization-based OCP acceleration and deceleration model is introduced, so that a lane changing strategy is more accurate and available; in the acceleration and deceleration lane changing strategy process, a two-layer structure of coarse screening and fine screening is arranged, the advantages that the calculation power consumption is small, and an acceleration and deceleration model finely meets the driving requirement are considered, meanwhile, processing of an extreme scene is further involved, and the safety of the whole lane changing process is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of autonomous driving, and relates to intelligent acceleration and deceleration lane changing for autonomous driving, and specifically to an intelligent acceleration and deceleration lane changing decision method for autonomous driving. Background Art

[0002] Lane-changing decision algorithms play a vital role in assisted driving and autonomous driving systems. They not only ensure that vehicles can move laterally safely and efficiently in a multi-lane environment, but also improve the efficiency of the overall traffic flow. By accurately evaluating the surrounding traffic conditions and predicting the behavior of other vehicles, lane-changing algorithms can intelligently select the best time and path to change lanes, thereby reducing potential collision risks and optimizing the vehicle's driving trajectory. In addition, it can also realize autonomous navigation of vehicles in complex traffic environments, providing drivers with a more comfortable and convenient driving experience. It is one of the key technologies for realizing highly automated and intelligent transportation systems.

[0003] Currently, the main lane-changing decision-making methods in the industry can be divided into four main categories: rule-based methods, model-based methods, game theory-based methods, and learning-based methods.

[0004] Rule-based lane changing methods are a traditional method for decision making in autonomous driving systems. This type of method guides the vehicle's lane changing behavior through a series of predefined logical rules. The following are some specific rule-based lane changing methods and their considerations, advantages and disadvantages:

[0005] 1. Lane changing rule based on speed difference: This rule considers the speed difference between the current vehicle speed and the vehicle speed in the target lane. If the vehicle speed in the target lane is higher than the current lane and the speed difference exceeds a certain threshold, the lane change is executed;

[0006] 2. Distance-based lane-changing rule: This rule considers the distance to the vehicle ahead and the available space in the target lane. If the distance to the vehicle ahead is large enough and there is enough space in the target lane, the lane change is allowed.

[0007] 3. Safety-based lane-changing rules: This rule takes into account the safe distance from other vehicles and the risk of collision during the lane-changing process, ensuring that a safe distance is maintained between the vehicles in front and behind the target lane during the lane-changing process to avoid collisions.

[0008] 4. Lane-changing rules based on driver intention: Based on the driver's intention, such as turn signal, lever signal, etc., when the driver's intention to change lanes is detected, the system automatically evaluates and executes the lane change;

[0009] 5. Lane changing based on global navigation rules: Based on the results of global navigation, such as the left and right turn lanes ahead and information such as on- and off-ramps, the lane change is performed at appropriate lane change points to ensure that the lane change decision is consistent with the global path planning.

[0010] The rule-based lane-changing decision method has the following advantages: 1) Simple and easy to understand: the algorithm is easy to understand and implement, and the rules are clear, which makes it easy for engineers and users to understand the working principle; 2) Low computational cost: such algorithms usually do not require complex calculations, can be executed in real time, and are suitable for implementation on resource-constrained systems; 3) Easy to adjust: new rules can be adjusted or added as needed to adapt to different driving scenarios and traffic regulations. At the same time, the rule-based lane-changing method also has many disadvantages, such as: 1) Poor flexibility and generalization: rule-based algorithms may have difficulty adapting to complex and changing traffic environments, and may not be able to make the best decisions for unforeseen situations; 2) Rule conflicts: For complex traffic flows and traffic regulations, different rules may conflict with each other, and additional logic is required to resolve these conflicts; 3) Difficult to optimize: rule-based algorithms are usually difficult to optimize to achieve the global optimal solution because they rely on local, static rules; at the same time, the performance of the algorithm is highly dependent on the design quality of the rules, and improperly designed rules may lead to undesirable lane-changing behavior.

[0011] The core principle of model-based lane changing methods is to use mathematical models to predict and simulate the behavior of vehicles on the road. These models usually consider the kinematics and dynamics of the vehicle, as well as the interaction between the vehicle and other vehicles and the environment. The model predicts the future state based on the current vehicle state and the perceived environmental information, and formulates a lane changing strategy accordingly. The main models related to driving and lane changing are:

[0012] 1. The IDM model, or Intelligent Driver Model, is a mathematical model used to simulate the following behavior of a vehicle on the road. It was proposed by German scientists Martin Treiber and Arne Kesting in 1999. The IDM model defines the speed difference formula, acceleration formula, safety distance formula, etc. based on driving safety and comfort to ensure the safety and stability of following the vehicle. Due to its simplicity and effectiveness, the IDM model has become one of the standard tools in the field of traffic flow simulation and vehicle dynamic modeling; however, the IDM model only considers the model of the vehicle following the front vehicle, and is not suitable for more complex traffic scenarios such as lane changing;

[0013] 2. The Gipps model was proposed by British traffic researcher Peter Gipps in the early 1980s to describe and predict the lane-changing behavior of vehicles on multi-lane roads. It calculates the probability of lane change based on the target lane distance difference and the speed-time relationship. The Gipps model is simple and intuitive, but too conservative and has limited adaptability to actual driving scenarios;

[0014] 3. The MOBIL model (Minimizing Overall Braking Induced by Lane-changes) is a vehicle lane-changing model that focuses on reducing the number and intensity of overall braking caused by lane changes, aiming to improve the stability and safety of traffic flow. The MOBIL model was proposed by Arne Kesting as a more detailed description of vehicle lane-changing behavior. The advantage of the MOBIL model is that it comprehensively considers the overall traffic flow to reduce braking and achieve a balance between safety and efficiency; the disadvantage is that the parameters are relatively sensitive and difficult to generalize, and the parameters need to be adjusted based on the specific traffic environment;

[0015] 4. Based on the optimal control model, the lane changing problem is defined as an optimal control problem. By minimizing the optimization objective and satisfying the kinematic and dynamic constraints, the lane changing timing and trajectory are obtained. The advantage of the lane changing model based on optimal control is that it can find the global optimal solution in the defined optimization space; the disadvantage is that it is difficult to establish an optimization problem model for scenes with a large number of traffic participants. The optimization objective or constraint of the model may be non-convex or nonlinear, which makes the problem difficult to solve and the calculation time is difficult to meet the needs of real-time decision-making and planning.

[0016] Model-based lane-changing methods, although there are a variety of different models that can cope with different driving scenarios and find the optimal solution for a specific scenario. However, these methods may be computationally complex, sensitive to model parameters, and require a lot of computing resources to ensure real-time decision-making in a dynamic traffic environment.

[0017] The lane-changing method based on game theory regards the interactive behavior of vehicles in a traffic environment as a game, in which each vehicle tries to make the best decision based on its own interests while taking into account the behavior of other vehicles. This method is often used to simulate and optimize decision-making problems in multi-agent systems, such as the lane-changing strategy of autonomous vehicles. The advantage of the lane-changing method based on game theory is that it can achieve the optimal lane-changing strategy for coordination between vehicles, improve the efficiency of the overall traffic flow, and has strong adaptability to complex dynamic traffic flows; the disadvantage is that the calculation complexity is high, and it is difficult to meet the requirements of real-time calculation at this stage.

[0018] Learning-based methods, especially lane-changing algorithms in the field of autonomous driving, usually involve machine learning or deep learning techniques to learn lane-changing strategies from data. The main learning-based lane-changing methods in academia are:

[0019] 1. Supervised learning-based methods: learning labeled data sets based on methods such as decision trees, support vector machines (SVMs), and deep neural networks to learn lane-changing behavior and timing;

[0020] 2. Reinforcement Learning (RL)-based methods: Based on Q-learning, SARSA, deep Q-network (DQN), policy gradient method, Actor-Critic method, etc., the lane-changing behavior and timing are learned by interacting with the environment to maximize the reward;

[0021] 3. Imitation Learning: Imitate the driver’s lane-changing behavior through a neural network combined with human expert driving data.

[0022] Learning-based lane-changing methods can adaptively learn complex lane-changing strategies from a large amount of data, and have good generalization capabilities and the ability to adapt to changing traffic environments. However, these methods rely on a large amount of training data, may require expensive computing resources, and still face challenges in ensuring the interpretability of the model and avoiding overfitting.

[0023] In summary, the various existing lane-changing decision methods have their own advantages and disadvantages, and it is impossible to use one method to cover various lane-changing requirements in different driving scenarios. Summary of the invention

[0024] In view of the above problems, the main purpose of the present invention is to design an intelligent acceleration and deceleration lane changing decision method for autonomous driving, so as to solve the problem that the existing lane changing methods have a narrow coverage and cannot adapt to various lane changing scenarios and various lane changing requirements.

[0025] In order to achieve the above purpose, the present invention adopts the following technical solutions:

[0026] An intelligent acceleration and deceleration lane-changing decision method for autonomous driving, the decision method uses a two-layer hierarchical architecture, namely a HighLevel layer and a Policy layer;

[0027] At the HighLevel layer, the upstream data is integrated to generate lane change intentions. The upstream data includes one or more of manual instructions, global navigation, road priority, and traffic efficiency.

[0028] At the level of lane change intention generation, an evaluation function is defined for different lanes, namely the lane reward function R. When generating the lane change intention, the lane with the largest R value is selected as the target lane.

[0029] At the policy layer, based on multiple models, lane-changing space and timing are selected to generate acceleration and deceleration lane-changing strategies.

[0030] According to actual driving scenarios, extreme scenarios are identified and handled, including sudden acceleration of the vehicle behind the target lane during lane changing, lane changing in construction areas, and dangerous lane changing caused by untimely upstream perception.

[0031] As a further description of the present invention, in the process of generating lane change intention, when changing lanes based on manual instructions, the lane reward function expression is:

[0032]

[0033] The command lane is the target lane.

[0034] As a further description of the present invention, in the process of generating lane change intention, when changing lanes based on global navigation, a Routing Reward function is designed, and the reward value of the i-th lane in the Routing dimension is The expression is:

[0035]

[0036] Among them, Routing Reward indicates the attractiveness of the lane near the operation point or the must-pass point to the vehicle; w i (d) = (1 + αr i ) -βd , represents the weight function of the i-th lane at the current distance d, d is the longitudinal distance from the current vehicle position to the operation point, r i is the indicator variable of whether the i-th lane has a work point, r i =1 means there is a work point in the i-th lane, r i = 0 means that there is no operation point in the i-th lane, α is the coefficient for controlling the influence weight of the operation point, β is the coefficient for controlling the distance attenuation, and p i is the priority or preference weight of the ith lane, p j is the priority or preference weight of the jth lane, and N is the total number of lanes in the horizontal direction at the current position of the vehicle;

[0037] Will Normalize to the interval [0,1] and select the lane corresponding to the maximum value as the target lane.

[0038] As a further description of the present invention, in the process of generating lane change intention, when changing lanes based on road priority, a reward function is defined, and the reward value in the lane priority dimension of the i-th lane is The expression is:

[0039]

[0040] Where x is the priority of the i-th lane, which is an integer between 0 and A. The larger the value, the higher the priority. 0 indicates an absolute no-entry zone, A indicates a completely recommended lane, k is the steepness of the function, x0 is the default lane priority, and e is the natural logarithm.

[0041] The above rewards based on priority need to change lanes to each lane, or Normalize to the interval [0,1] and select the lane corresponding to the maximum value as the target lane.

[0042] As a further description of the present invention, in the process of generating lane change intention, when changing lanes based on traffic efficiency, the traffic speeds of different lanes are considered, and the lane traffic efficiency Reward function is expressed as:

[0043]

[0044]

[0045] Among them, E i is the lane traffic efficiency, is the reward value of the lane priority dimension of the i-th lane, v d is the preset speed limit value of the lane, v ego is the current speed of the vehicle, v target is the speed of the obstacle vehicle in front of the target lane, E i With E j are the reward values ​​of the efficiency dimension of the i-th and j-th lanes, respectively, and e is the natural logarithm;

[0046] Select The lane corresponding to the maximum value of is the target lane.

[0047] As a further description of the present invention, when two or more of the manual instructions, global navigation, road priority, and lane change of traffic efficiency are used in combination, the Reward function expression is:

[0048] R=w o R o +w r R r +w p R p +w e R e ;

[0049] Among them, w is the weight of the reward function in different dimensions, R o , R r , R p , R e are the reward values ​​corresponding to manual instructions, global navigation, road priority, and traffic efficiency, respectively, o 、w r 、w p 、w e Corresponding to R o , R r , R p , R e The weight of

[0050] The lane corresponding to the maximum value of R is selected as the target lane.

[0051] As a further description of the present invention, the acceleration and deceleration lane changing strategy includes: candidate lane changing space Gap generation, lane changing space Gap coarse screening, and lane changing space Gap fine screening, wherein the candidate lane changing space Gap generation includes target lane obstacle projection and calculation and storage of all candidate lane changing spaces;

[0052] The process of projecting the target lane obstacle is as follows: according to the lane information and obstacle information, the Frenet coordinate system is used to project the target lane obstacle along the reference line s at each moment;

[0053] The process of calculating and saving all candidate lane change spaces is as follows: traversing the target lane, identifying all lane change gaps, and determining the data structure of each lane change gap, including the reference line ID, the upper bound of the gap, and the lower bound of the gap;

[0054] For each time step, the dynamic state of the lane changing space Gap in time is recorded, and the information of the lane changing space Gap is saved.

[0055] As a further description of the present invention, the lane changing space Gap rough screening includes: screening based on the distance from the current vehicle to the lane changing space Gap, screening based on the space size of the lane changing space Gap, screening based on the speed check of the vehicle behind the lane changing space Gap, and screening based on the speed check of the vehicle in front of the lane changing space Gap;

[0056] The screening based on the distance from the current vehicle to the lane-changing space Gap is to filter out the lane-changing space Gap whose distance from the vehicle is greater than the set threshold. The expression is:

[0057] D=max(v ego T,d min );

[0058] Where D is the threshold for lane-changing space screening, v ego is the current vehicle speed, T is the average lane changing time parameter, d min is the minimum distance threshold of the lane-changing space. When the distance from the vehicle to the lane-changing space exceeds D, the lane-changing space is discarded.

[0059] The screening based on the size of the lane changing space Gap is as follows: setting a threshold min_gap_dis of the size of the lane changing space, and filtering out the lane changing space that is smaller than the threshold min_gap_dis;

[0060] The screening based on the speed check of the rear vehicle in the lane changing space Gap is as follows: for the check when the rear vehicle in the lane changing space is close to the vehicle, when the speed of the rear vehicle is greater than the speed of the vehicle plus the allowed speed difference, the lane changing space is screened out;

[0061] The screening based on the lane changing space Gap front vehicle speed check is as follows: for the check when the front vehicle in the lane changing space is close to the own vehicle, if the front vehicle speed is less than the own vehicle speed minus the speed difference allowed to be reduced, the lane changing space is screened out.

[0062] As a further description of the present invention, the fine screening of the lane change space Gap includes the following steps:

[0063] The safe lane-changing condition is defined as the instantaneous safety condition and the safety window is not less than the set threshold. The instantaneous safety condition is that at time t0, the distance behind the vehicle is greater than d1, and the distance in front is greater than d2, then it is considered instantaneous safety. The safety window is based on the speed curve of the vehicle, the speed curves of the front and rear vehicles. At any time, the front and rear vehicles maintain the set safety distance with the vehicle, and the set safety distance is maintained for t consecutive moments thereafter, then it is considered that there is a safety window.

[0064] Based on the candidate lane-changing space, the velocity profile of the ego vehicle is generated as:

[0065] Define the velocity differential model, the model expression is:

[0066]

[0067] Among them, s is the current travel distance of the vehicle, v is the speed, a is the acceleration, and j is the acceleration derivative jerk;

[0068] Define the optimization goal, the vehicle reaches the state within the upper and lower bounds of the lane change space Gap, and the optimization objective function expression is:

[0069]

[0070] Among them, w s , w v , w a , w j are the weight parameters of s, v, a, and jerk respectively. sr is the reference driving distance, which is set to sr = 0.5 (start_s + end_s), that is, the expected s of the ego vehicle is the middle position of the lane change space;

[0071] In the time range from t0 to tN, considering the constraints of maximum acceleration, maximum and minimum jerk, a quadratic optimization problem is constructed and solved to obtain the velocity curve of the ego vehicle;

[0072] For each lane-changing space, the solver calculates the time t required for its speed curve to reach the safe lane-changing condition, and selects the lane-changing space with the smallest t as the optimal lane-changing space. Based on the speed curve corresponding to the optimal lane-changing space, the lane-changing timing and the acceleration and deceleration model of the vehicle to reach the appropriate timing are obtained, and the lane-changing strategy is generated to output the lane-changing trajectory.

[0073] As a further description of the present invention, in actual driving scenarios, extreme scenarios are identified and processed. Among them, when a dangerous lane change is caused by a sudden acceleration of the vehicle behind in the target lane and untimely upstream perception, the vehicle cancels the current lane changing behavior; when changing lanes in a construction area, the vehicle triggers the lane changing intention before approaching the construction area, and defines a lane risk model based on the obstacle status of the lane and the construction area. The lane risk model is combined at the lane changing intention generation level to trigger the lane changing to avoid the construction area.

[0074] Compared with the prior art, the technical effects of the present invention are:

[0075] The present invention provides an automatic driving intelligent acceleration and deceleration lane changing decision method, including three processes: lane changing intention generation, acceleration and deceleration lane changing strategy, and extreme scene recognition and processing. Different from the traditional single method, it is a hybrid method, which combines the advantages of simplicity and good interpretability of rule-based methods, and introduces an efficient OCP acceleration and deceleration model based on optimization, so that the lane changing strategy is more accurate and usable. In the process of acceleration and deceleration lane changing strategy, a two-layer structure of coarse screening and fine screening is set, which takes into account the advantages of low computing power consumption and the acceleration and deceleration model that precisely fits driving needs. At the same time, it also involves the processing of extreme scenes to improve the safety of the overall lane changing process. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 It is a schematic diagram of the overall method architecture of the present invention;

[0077] Figure 2 A logical schematic diagram of the lane changing method of the present invention;

[0078] Figure 3 A schematic diagram is generated for a candidate lane-changing space Gap of the present invention;

[0079] Figure 4 This is a schematic diagram of the distance from the current vehicle to the lane-changing space Gap of the present invention;

[0080] Figure 5 A schematic diagram of a construction area scenario and a lane risk model of the present invention;

[0081] Figure 6 Schematic diagram of the cloud dynamic solution of the present invention. DETAILED DESCRIPTION

[0082] The present invention is described in detail below in conjunction with the accompanying drawings:

[0083] In one embodiment of the present invention, an automatic driving intelligent acceleration and deceleration lane change decision method is disclosed, referring to Figure 1-6 As shown, the decision-making method uses a two-layer hierarchical architecture, namely the HighLevel layer and the Policy layer;

[0084] At the HighLevel layer, the upstream data is integrated to generate lane change intentions. The upstream data includes one or more of manual instructions, global navigation, road priority, and traffic efficiency.

[0085] At the level of lane change intention generation, an evaluation function is defined for different lanes, namely the lane reward function R. When generating the lane change intention, the lane with the largest R value is selected as the target lane.

[0086] At the policy layer, based on multiple models, lane-changing space and timing are selected to generate acceleration and deceleration lane-changing strategies.

[0087] According to actual driving scenarios, extreme scenarios are identified and handled, including sudden acceleration of the vehicle behind the target lane during lane changing, lane changing in construction areas, and dangerous lane changing caused by untimely upstream perception.

[0088] Specifically, in this embodiment, the above decision-making method is explained from three aspects: lane change intention generation, acceleration and deceleration lane change strategy, and extreme scene recognition and processing.

[0089] 1. Lane Change Intention Generation

[0090] In order to make the lane change behavior explainable and meet the driving needs of different scenarios, the lane change intention generation layer mainly uses a rule-based solution to decide whether to initiate a lane change and the target lane selection for the lane change. In order to better describe the lane change intention, this embodiment defines a set of standard and unified evaluation functions for different lanes, which is called the lane reward function R. By calculating and comparing the reward values ​​R of all lanes, the lane with the largest R value is selected as the target lane for lane change.

[0091] Specifically, in this embodiment, the R value of the lane change map is obtained for the upstream data: manual instructions, global navigation, road priority, and traffic efficiency.

[0092] 1. Manual command lane change

[0093] Manually instructed lane changing usually refers to lane changing with manual intervention or forced control of traffic scenes. The main types include lane changing by the driver's lever, lane changing by remote control system, and lane changing by pulling over when fault degradation occurs. This type of lane changing intention is initiated manually or by clear forced control rules; specifically, in the process of generating lane changing intention, based on manually instructed lane changing, the lane reward function expression is:

[0094]

[0095] The command lane is the target lane.

[0096] 2. Lane change based on global navigation (routing)

[0097] The purpose of routing lane change is to pass through a certain waypoint in order to complete a certain task, such as changing lanes to a certain lane when going down the ramp, or changing lanes to a certain lane in order to complete a certain operation (such as getting on and off passengers or loading and unloading goods).

[0098] It is related to 4 factors:

[0099] 1) The longitudinal distance d between the vehicle and the necessary passage point (operation point);

[0100] 2) Whether the target lane is routing (whether the target lane has an operation point) r, the value of r is 0 or 1;

[0101] 3) The current speed v of the vehicle will directly affect the distance s1 of a single lane change;

[0102] 4) The number of lanes l of the lane where the current vehicle must pass through the waypoint;

[0103] Specifically, in this embodiment, during the process of generating lane change intention, based on the global navigation lane change, a RoutingReward function is designed, and the reward value of the i-th lane in the routing dimension is The expression is:

[0104]

[0105] Among them, Routing Reward indicates the attractiveness of the lane near the operation point or the must-pass point to the vehicle; w i (d) = (1 + αr i ) -βd , represents the weight function of the i-th lane at the current distance d, d is the longitudinal distance from the current position of the vehicle to the working point. The above formula makes the desire to change lanes to the working lane stronger as the vehicle approaches the lane-changing point. i is the indicator variable of whether the i-th lane has a work point, r i =1 means there is a work point in the i-th lane, r i = 0 means that there is no operation point in the lane, α is the coefficient for controlling the influence weight of the operation point, β is the coefficient for controlling distance attenuation, p i is the priority or preference weight of the ith lane, p jis the priority or preference weight of the jth lane, and N is the total number of lanes in the lateral direction of the current vehicle location.

[0106] Assuming that the distance required to complete a single lane change is s1, and the number of lateral lanes between the lane where the current vehicle is located and the lane where the operation point is located is n, the lane change must be initiated before d = ns1. Therefore, when d>ns1, the above function needs to be truncated, that is, the Routing Reward is directly set to 1, or it can be designed as a function about n1 based on a continuous function.

[0107] Will Use the softmax function to normalize to the [0,1] interval, and select the lane corresponding to the maximum value as the target lane.

[0108] 3. Lane change based on road priority

[0109] In addition to the above-mentioned command intent and routing intent, in the lane-changing scenario of autonomous driving, different lane priorities need to be considered to express which lanes are usually more recommended and which lanes are not recommended or prohibited. For example, the emergency lane on the highway should not be used normally, but it can be used in an emergency; in the operation lane and driving lane in the port, most of the non-operation time, it is preferred to use the driving lane rather than occupy the operation lane. These requirements can be expressed through the lane priority attribute.

[0110] Specifically, in this embodiment, during the process of generating lane change intention, when changing lanes based on road priority, a reward function is defined, and the reward value in the lane priority dimension of the i-th lane is The expression is:

[0111]

[0112] Where x is the priority of the i-th lane, which is an integer between 0 and A. The larger the value, the higher the priority. 0 means absolutely no lane, A means a completely recommended lane, k is the steepness of the function, usually 0.2, x0 is the default lane priority, e is the natural logarithm, at x = x0, the lane passing reward is 0.5. The above expression is based on the reward for changing lanes to each lane based on the priority, and can also be replaced by Use the softmax function to normalize to the [0,1] interval, and select the lane corresponding to the maximum value as the target lane.

[0113] In this embodiment, the above-mentioned A is set to 100, and the priority is an integer between 0-100. The larger the number, the higher the priority, and the smaller the number, the lower the priority. 0 means an absolute no-entry lane, and 100 means a completely recommended lane. Under normal circumstances, the priority of an ordinary lane is 50, the recommended lane is 60, the work lane is 30, and the emergency lane is 10.

[0114] 4. Lane change based on traffic efficiency

[0115] Lane changing based on traffic efficiency mainly considers the traffic speed of different lanes. The vehicle always tends to choose a faster lane to maximize efficiency. Traffic efficiency is mainly related to the current speed of the vehicle, the speed of the target lane, and the speed limit of the target lane.

[0116] Specifically, in this embodiment, during the process of generating the lane change intention, when changing lanes based on the traffic efficiency, the traffic speeds of different lanes are taken into consideration, and the lane traffic efficiency reward function is expressed as:

[0117]

[0118] Among them, E i is the lane traffic efficiency, is the reward value Rip of the lane priority dimension of the i-th lane, v d is the preset speed limit value of the lane, v ego is the current speed of the vehicle, v target is the speed of the obstacle vehicle in front of the target lane, E i With E j are the efficiency dimension reward values ​​of the i-th and j-th lanes respectively, and e is the natural logarithm; The lane corresponding to the maximum value of is the target lane.

[0119] 5. Use two or more types in combination

[0120] The above four lane-changing intentions in different dimensions may be used individually or in combination in different driving scenarios. Therefore, the final lane comprehensive reward is defined based on the above four lane-changing intentions.

[0121] Specifically, in this embodiment, when two or more of the lane changing options of manual instruction, global navigation, road priority, and traffic efficiency are used in combination, the Reward function expression is:

[0122] R=w o R o +w r R r +w p R p +w e R e ;

[0123] Among them, w is the weight of the reward function in different dimensions. Different weight parameters can be used in different lane changing scenarios. o , R r , R p , R eare the reward values ​​corresponding to manual instructions, global navigation, road priority, and traffic efficiency, respectively, o 、w r 、w p 、w e Corresponding to R o , R r , R p , R e The weight of R is selected; the lane corresponding to the maximum value of R is selected as the target lane, and then it can be decided whether to change lanes and which lane is the target lane.

[0124] 2. Acceleration and deceleration lane changing strategy

[0125] The lane changing intention mentioned above partially solves the problem of whether to change lanes and the selection of the target lane for lane changing. After the lane changing intention is generated, it is necessary to specifically consider the timing of lane changing and adjust to the appropriate speed to find a safe lane changing space in the target lane for lane changing. In this embodiment, the lane changing space is defined as Gap (representing a longitudinal space of the target lane). The specific lane changing strategy includes three main steps: candidate Gap generation, Gap coarse screening and Gap fine screening process.

[0126] 1. Candidate Gap Generation

[0127] The process of generating candidate gaps requires generating all candidate lane change space gaps based on the distribution of obstacle vehicles in the target lane.

[0128] Specifically, in this embodiment, the candidate lane change space Gap generation includes: 1) target lane obstacle projection; 2) calculating and saving all candidate lane change spaces; wherein the process of target lane obstacle projection is: according to lane information and obstacle information, the Frenet coordinate system is used to project the target lane obstacle along the reference line s direction at each moment; the above lane information includes the center line of the lane, lane width data, boundary type, etc., and the above obstacle information includes the obstacle's current position, obstacle speed information, predicted trajectory for a period of time in the future, etc.;

[0129] The process of calculating and saving all candidate lane-changing spaces is as follows: traverse the target lane, identify all lane-changing space gaps, and determine the data structure of each lane-changing space gap. The gap data structure includes: 1) reference line ID: the reference line represents the lane centerline curve after smoothing operation; 2) upper bound of the gap: the forward vehicle ID of the gap space, which can be used to query the vehicle's motion status at each moment according to the ID. If there is no vehicle in the forward direction, the ID is empty; 3) lower bound of the gap: the backward vehicle ID of the gap space, which can be used to query the vehicle's motion status at each moment according to the ID. If there is no vehicle in the backward direction, the ID is empty; 4) dynamic state of the gap in time: for each time step, record the dynamic state of the lane-changing space gap in time, and save the information of the lane-changing space gap, including the projection of the obstacles in front and behind the gap, speed, and acceleration (corresponding to s, v, a of ST) at each time step.

[0130] like Figure 3 As shown in the figure, the candidate lane change space Gap generation process is as follows: the ego vehicle is in the rightmost lane, the target lane is the left lane, there are two vehicles in the target lane, and their IDs are 2 and 4 respectively. Based on the above projection method, three Gap spaces shown by the dotted arrows in the figure can be obtained: Gap1: the upper bound is empty, and the lower bound is vehicle 2; Gap2: the upper bound is vehicle 2, and the lower bound is vehicle 4; Gap3: the upper bound is vehicle 4, and the lower bound is empty.

[0131] 2. Gap coarse screening strategy

[0132] The above steps generate candidate gap spaces for the target lane. In this embodiment, the gap spaces that are obviously unreasonable or unreachable are subsequently screened out based on a simple rule-based algorithm. This process is called gap coarse screening. Four rules are defined as follows to screen the gaps, namely: screening based on the distance from the current vehicle to the lane changing space gap, screening based on the size of the lane changing space gap, screening based on the speed check of the vehicle behind the lane changing space gap, and screening based on the speed check of the vehicle in front of the lane changing space gap.

[0133] (1) Filtering based on the distance between the current vehicle and the lane change space Gap

[0134] The current distance from the vehicle to Gap can be divided into three cases, such as Figure 4As shown in the figure, the first case is: the current position of the vehicle is behind the vehicle in front of the gap and in front of the vehicle behind the gap, that is, font_dist and back_dist in the figure below are both greater than 0, then the distance from the vehicle to the lane changing space Gap is 0; the second case is: the gap is in front of the vehicle, then the distance from the vehicle to the lane changing space Gap is defined as the s-coordinate of the lower boundary of the Gap minus the s-coordinate of the rear of the vehicle; the third case is: the gap is behind the vehicle, then the distance from the vehicle to the lane changing space Gap is defined as the s-coordinate of the front of the vehicle minus the s-coordinate of the upper boundary of the Gap.

[0135] After defining the distance from the current vehicle to the gap, the lane change gaps whose distance from the vehicle is greater than the set threshold are filtered out based on the rules (gaps that are too far away usually require a long time and large acceleration and deceleration to reach a safe state). The screening logic is:

[0136] D=max(v ego T,d min );

[0137] Where D is the threshold for lane-changing space screening, v ego is the current vehicle speed, T is the average lane change time parameter, usually 8 seconds, and d min is the minimum distance threshold of the lane-changing space. When the vehicle is at an extremely low speed, D is guaranteed to be valid. When the distance from the vehicle to the lane-changing space exceeds D, the lane-changing space Gap will be discarded in the coarse screening stage.

[0138] (2) Screening based on the size of the lane-changing gap

[0139] The distance between the upper boundary s coordinate and the lower boundary s coordinate of the gap in the longitudinal space shall not be less than the vehicle length. At the same time, in high-speed scenarios, the gap space needs to be further enlarged to ensure safety. In this embodiment, a threshold min_gap_dis for the size of the lane change space is set, and the lane change space that is less than the threshold min_gap_dis is filtered out;

[0140] (3) Screening based on the speed check of the vehicle behind in the lane change gap

[0141] For the rear vehicle in the lane change space Gap, when the distance to the ego vehicle is relatively close, the lane change space is screened out when the speed of the rear vehicle is greater than the ego vehicle speed plus the speed difference allowed to exceed. Specifically, according to different driving conditions, based on the collision time judgment, or the distance determined by the IDM following model, if the rear vehicle is close to the ego vehicle, the speed should not exceed a certain threshold of the ego vehicle speed, that is, the Gap of follower_v>ego_v+back_speed_offset is discarded, where follower_v is the speed of the rear vehicle in the Gap, ego_v is the ego vehicle speed, and back_speed_offset is the defined speed range that is allowed to exceed the ego vehicle, and this value should not be too large. This check can be ignored for vehicles that are far behind (which can be judged by relative distance and relative speed).

[0142] (4) Screening based on the lane change gap and the speed check of the vehicle ahead

[0143] The speed check of the preceding vehicle is similar to the speed check of the following vehicle. For the check when the preceding vehicle is close to the ego vehicle in the lane-changing space, if the preceding vehicle speed is less than the ego vehicle speed minus the speed difference allowed to be reduced, the lane-changing space is screened out. Specifically, according to different driving conditions, based on the collision time judgment, or the distance determined by the IDM following model, the speed of the preceding vehicle at a closer distance must not be lower than the set threshold of the ego vehicle speed, and the check is ignored for the vehicle far ahead. Gaps that do not satisfy front_v>=ego_v-front_speed_offset will be discarded, where front_v is the speed of the preceding vehicle in the gap, ego_v is the ego vehicle speed, and front_speed_offset is the speed difference allowed to reduce the defined speed, which is a set threshold.

[0144] It should be noted that, in this embodiment, the above-mentioned distance checks between the rear vehicle and the vehicle itself, and between the front vehicle and the vehicle itself, are differentiated according to high-speed and low-speed scenarios, and are not limited to fixed distance lengths; usually closer checks are defined by time intervals, such as a 2 to 3 second overtaking time, which is usually more than 50 meters in high-speed scenarios and reduced to 2-3 meters in extremely low-speed congestion scenarios.

[0145] 3. Gap screening strategy

[0146] Based on the above-mentioned gap coarse screening strategy, the candidate gaps that are obviously unreachable or do not meet the kinematic constraints can be screened out. However, in complex traffic flow scenarios, there are usually a few gaps that meet the conditions after the coarse screening. At this time, further screening strategies are required to select better gaps to perform operations. This process is called gap fine screening.

[0147] In the fine screening stage, it is necessary to solve the vehicle speed curve that meets the safe lane changing conditions based on the candidate lane changing space gaps, and then select the gap with the shortest time and the most comfortable acceleration and deceleration to actually perform the lane changing based on the speed curve. During the fine screening process, the acceleration and deceleration curves of the vehicle are also solved and can be directly used for the subsequent lane changing trajectory generation.

[0148] First, it is necessary to define what is a safe lane-changing condition. Different from the traditional method that only considers the instantaneous front and rear vehicle distance, it also calculates the safety conditions within a future time window based on the deduced speed curve of the vehicle, which is called a safety window. In this embodiment, the safe lane-changing condition is defined as having instantaneous safety conditions, and the safety window is not less than a set threshold (for example, 3 seconds); the instantaneous safety condition is that at time t0, the distance behind the vehicle is greater than d1, and the distance in front is greater than d2, then it is considered instantaneous safety; the safety window is based on the speed curve of the vehicle, the speed curves of the front and rear vehicles, at any time, the front and rear vehicles maintain a set safety distance from the vehicle, and the set safety distance is maintained for t consecutive moments thereafter, then it is considered that there is a safety window.

[0149] Based on the candidate lane-changing space, the speed curve of the vehicle is generated in a method, which is different from the traditional IDM, PID and other models. In this embodiment, an OCP intelligent acceleration and deceleration speed model is proposed based on the optimal control theory.

[0150] First, define the velocity differential model, which is expressed as:

[0151]

[0152] Where s is the current driving distance of the vehicle, v is the speed, a is the acceleration, and j is the acceleration derivative jerk. In this embodiment, the purpose is to make the vehicle reach the state within the upper and lower bounds of the lane change space Gap as quickly as possible on the basis of making the jerk and acceleration as smooth as possible. Therefore, the optimization goal is defined as:

[0153]

[0154] Among them, w s , w v , w a , w j are the weight parameters of s, v, a, and jerk respectively, and sr is the reference driving distance. In order to achieve the state of the ego vehicle at the upper and lower boundaries of the Gap, in this embodiment, it is set to sr=0.5(start_s+end_s), that is, the expected s of the ego vehicle is the middle position of the lane change space.

[0155] In addition to the above differential model, within the time range of t0 to tN, considering the constraints of maximum acceleration, maximum and minimum jerk, a quadratic optimization problem is constructed, and the problem is solved to obtain the speed curve of the ego vehicle. At the same time, for each lane-changing space, based on the speed curve of the ego vehicle and the speed curves of the vehicles before and after the Gap, the solver is used to solve the time t required for its speed curve to reach the safe lane-changing condition, and the lane-changing space with the smallest t is selected as the optimal lane-changing space. Based on the speed curve corresponding to the optimal lane-changing space, the lane-changing timing and the acceleration and deceleration model of the ego vehicle at the appropriate timing are obtained, and the lane-changing strategy is generated to output the lane-changing trajectory.

[0156] The core of the above fine screening step is to solve the speed curve of the vehicle based on the OCP optimization model for each candidate gap left after the rough screening, calculate the time t required for the speed curve of each gap to reach the safe lane change condition, and finally select the gap with the smallest t as the final optimal gap. At the same time, based on the speed curve corresponding to the optimal gap, the appropriate lane change timing can be obtained, as well as the acceleration and deceleration model required by the vehicle to reach the appropriate timing.

[0157] 3. Extreme Scene Identification and Processing

[0158] Based on the above-mentioned Gap coarse screening and fine screening, the optimal Gap is selected from multiple candidate Gap, and a suitable acceleration and deceleration curve is generated. At this point, most lane change scenarios have been met. However, in actual driving scenarios, some extreme scenarios will also be encountered. In this embodiment, steps for identifying and processing extreme scenarios are added to make the entire lane changing method more reliable. Among them, when the rear vehicle in the target lane suddenly accelerates (the time distance is shortened to below the safety threshold, such as less than 3 seconds in high-speed scenarios) and when the upstream perception is not timely, resulting in dangerous lane changes, the vehicle cancels the current lane change behavior (lane change withdrawal); when changing lanes in the construction area, the vehicle triggers the lane change intention before approaching the construction area, and defines the lane risk model according to the obstacle status of the lane and the construction area. The lane risk model is combined at the lane change intention generation level to trigger the lane change to avoid the construction area.

[0159] 1) Lane change withdrawal

[0160] Lane change withdrawal means that at a certain moment, the self-vehicle determines that lane change is safe based on the current environment, but the lane change process becomes unsafe due to unstable environmental perception or aggressive and dangerous driving of other vehicles. In this case, the self-vehicle cancels the current lane change behavior and returns to the original lane while maintaining vehicle stability.

[0161] A typical lane change withdrawal scenario is: the ego vehicle is safe at first, but may collide with the vehicle coming from behind due to perception loss. Based on this, our lane change method incorporates a reliable lane change withdrawal processing strategy.

[0162] 2) Construction area treatment

[0163] Typical construction scenarios such as Figure 5 As shown, the vehicle needs to trigger the correct lane change intention to avoid the construction cones before approaching the construction area. Specifically, in this embodiment, a lane risk model is defined based on the positional relationship between the lane and the cones. The lane that is not affected by the cones is a low-risk lane, the lane that is partially invaded by the cones but is still large enough for vehicles to pass is a medium-risk lane, and the lane that is completely invaded by the cones and does not have enough space for vehicles to pass is a high-risk lane. The lane risk model can be combined with the lane change intention level to correctly trigger the lane change to avoid the construction area.

[0164] In addition, during high-speed driving, since the cones are small targets, the sensor's perception range is limited. Often, when the cones are sensed, the vehicle is already very close to the construction area, which will trigger a large deceleration and may even cause the vehicle behind to rear-end the vehicle. Based on this, in the specific implementation of this embodiment, a solution based on cloud-based dynamic layers can also be proposed.

[0165] Cloud dynamic layer solutions, such as Figure 6 As shown in the figure, in a multi-vehicle autonomous driving system, after the first vehicle passing through the construction area recognizes the construction area, it reports the area to the cloud scheduling system. The cloud system then transmits the construction area data back to other rear vehicles. The rear vehicles can then sense the existence of the construction area beyond visual range, thereby initiating lane changes in advance to avoid the construction area and avoid speed reduction or even sudden braking.

[0166] Through the above content, the technical solution of the present invention is disclosed. Compared with the prior art, the present invention has the following advantages:

[0167] 1. The lane-changing method of the present invention is different from the traditional single method. It is a hybrid method that combines the advantages of simplicity and good interpretability of the rule-based method, and introduces an efficient optimization-based OCP acceleration and deceleration model to make the lane-changing strategy more accurate and usable;

[0168] 2. In the step of selecting lane-changing space, the present invention designs a two-layer structure of coarse screening and fine screening, taking into account the advantages of low computing power consumption and the acceleration and deceleration model being finely tailored to driving needs.

[0169] 3. The present invention also involves the processing of extreme scenarios to improve the safety of the overall lane changing process.

[0170] The above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Other modifications or equivalent substitutions made to the technical solution of the present invention by ordinary technicians in this field should be included in the scope of the claims of the present invention as long as they do not depart from the spirit and scope of the technical solution of the present invention.

Claims

1. An intelligent acceleration and deceleration lane-changing decision method for autonomous driving, characterized in that: This decision-making method uses a two-layer hierarchical architecture, namely the HighLevel layer and the Policy layer; At the HighLevel layer, the upstream data is integrated to generate lane change intentions. The upstream data includes one or more of manual instructions, global navigation, road priority, and traffic efficiency. At the level of lane change intention generation, an evaluation function is defined for different lanes, namely the lane reward function R. When generating the lane change intention, the lane with the largest R value is selected as the target lane. At the policy layer, based on multiple models, lane-changing space and timing are selected to generate acceleration and deceleration lane-changing strategies. According to actual driving scenarios, extreme scenarios are identified and handled, including sudden acceleration of the vehicle behind the target lane during lane changing, lane changing in construction areas, and dangerous lane changing caused by untimely upstream perception.

2. The method for intelligent acceleration and deceleration lane changing decision-making for automatic driving according to claim 1, characterized in that: In the process of generating lane change intention, when changing lanes based on manual instructions, the lane reward function expression is: The command lane is the target lane.

3. The method for intelligent acceleration and deceleration lane changing decision-making for automatic driving according to claim 1, characterized in that: In the process of lane change intention generation, when changing lanes based on global navigation, the Routing Reward function is designed. The reward value of the i-th lane The expression is: Among them, Routing Reward indicates the attractiveness of the lane near the operation point or the must-pass point to the vehicle; w i (d) = (1 + αr i ) -βd , represents the weight function of the i-th lane at the current distance d, d is the longitudinal distance from the current vehicle position to the operation point, r i is the indicator variable of whether the i-th lane has a work point, r i =1 means there is a work point in the i-th lane, r i = 0 means that there is no operation point in the i-th lane, α is the coefficient for controlling the influence weight of the operation point, β is the coefficient for controlling the distance attenuation, and p i is the priority or preference weight of the ith lane, p i is the priority or preference weight of the jth lane, and N is the total number of lanes in the horizontal direction at the current position of the vehicle; Will Normalize to the interval [0, 1] and select the lane corresponding to the maximum value as the target lane.

4. The method for intelligent acceleration and deceleration lane changing decision-making for automatic driving according to claim 1, characterized in that: In the process of generating lane change intention, when changing lanes based on road priority, the Reward function is defined, and the reward value in the lane priority dimension of the i-th lane is The expression is: Where x is the priority of the i-th lane, which is an integer between 0 and A. The larger the value, the higher the priority. 0 indicates an absolute no-entry zone, A indicates a completely recommended lane, k is the steepness of the function, x0 is the default lane priority, and e is the natural logarithm. The above rewards based on priority need to change lanes to each lane, or Normalize to the interval [0, 1] and select the lane corresponding to the maximum value as the target lane.

5. The method for intelligent acceleration and deceleration lane changing decision-making for automatic driving according to claim 1, characterized in that: In the process of generating lane change intention, when changing lanes based on traffic efficiency, the traffic speeds of different lanes are considered, and the lane traffic efficiency reward function is expressed as: Among them, E i is the lane traffic efficiency, is the reward value of the lane priority dimension of the i-th lane, v d is the preset speed limit value of the lane, v ego is the current speed of the vehicle, v target is the speed of the obstacle vehicle in front of the target lane, E i With E j are the reward values ​​of the efficiency dimension of the i-th and j-th lanes, respectively, and e is the natural logarithm; Select The lane corresponding to the maximum value of is the target lane.

6. The method for intelligent acceleration and deceleration lane changing decision-making for automatic driving according to claim 1, characterized in that: When manual instructions, global navigation, road priority, and lane change of traffic efficiency are used in combination, the Reward function expression is: R=w o R o +w r R r +w p R p +w e R e ; Among them, w is the weight of the reward function in different dimensions, R o , R r , R p , R e are the reward values ​​corresponding to manual instructions, global navigation, road priority, and traffic efficiency, respectively, o 、w r 、w p 、w e Corresponding to R o , R r , R p , R e Weight: The lane corresponding to the maximum value of R is selected as the target lane.

7. The method for intelligent acceleration and deceleration lane changing decision-making for automatic driving according to claim 1, characterized in that: The acceleration and deceleration lane changing strategy includes: candidate lane changing space gap generation, lane changing space gap coarse screening, and lane changing space gap fine screening. The candidate lane changing space gap generation includes target lane obstacle projection and calculation and storage of all candidate lane changing spaces; The process of projecting the target lane obstacle is as follows: according to the lane information and obstacle information, the Frenet coordinate system is used to project the target lane obstacle along the reference line s at each moment; The process of calculating and saving all candidate lane change spaces is as follows: traversing the target lane, identifying all lane change gaps, and determining the data structure of each lane change gap, including the reference line ID, the upper bound of the gap, and the lower bound of the gap; For each time step, the dynamic state of the lane changing space Gap in time is recorded, and the information of the lane changing space Gap is saved.

8. The method for intelligent acceleration and deceleration lane-changing decision-making for automatic driving according to claim 7, characterized in that: The lane changing space gap rough screening includes: screening based on the distance from the current vehicle to the lane changing space gap, screening based on the space size of the lane changing space gap, screening based on the speed check of the vehicle behind the lane changing space gap, and screening based on the speed check of the vehicle in front of the lane changing space gap; The screening based on the distance from the current vehicle to the lane-changing space Gap is to filter out the lane-changing space Gap whose distance from the vehicle is greater than the set threshold. The expression is: D=max(v ego T,d min ); Where D is the threshold for lane-changing space screening, v ego is the current vehicle speed, T is the average lane changing time parameter, d min is the minimum distance threshold of the lane-changing space. When the distance from the vehicle to the lane-changing space exceeds D, the lane-changing space is discarded. The screening based on the size of the lane-changing space Gap is as follows: a threshold of the lane-changing space size min_gap_dis is set, and the lane-changing space that is smaller than the threshold min_gap_dis is screened out. The screening based on the speed check of the rear vehicle in the lane changing space Gap is as follows: for the check when the rear vehicle in the lane changing space is close to the vehicle, when the speed of the rear vehicle is greater than the speed of the vehicle plus the allowed speed difference, the lane changing space is screened out; The screening based on the lane changing space Gap front vehicle speed check is as follows: for the check when the front vehicle in the lane changing space is close to the own vehicle, if the front vehicle speed is less than the own vehicle speed minus the speed difference allowed to be reduced, the lane changing space is screened out.

9. The method for intelligent acceleration and deceleration lane-changing decision-making for automatic driving according to claim 8, characterized in that: The fine screening of the lane change space Gap includes the following steps: The safe lane-changing condition is defined as the instantaneous safety condition and the safety window is not less than the set threshold. The instantaneous safety condition is that at time t0, the distance behind the vehicle is greater than d1, and the distance in front is greater than d2, then it is considered instantaneous safety. The safety window is based on the speed curve of the vehicle, the speed curves of the front and rear vehicles. At any time, the front and rear vehicles maintain the set safety distance with the vehicle, and the set safety distance is maintained for t consecutive moments thereafter, then it is considered that there is a safety window. Based on the candidate lane-changing space, the velocity profile of the ego vehicle is generated as: Define the velocity differential model, the model expression is: Among them, s is the current travel distance of the vehicle, v is the speed, a is the acceleration, and j is the acceleration derivative jerk; Define the optimization goal, the vehicle reaches the state within the upper and lower bounds of the lane change space Gap, and the optimization objective function expression is: Among them, w s , w v , w a , w j are the weight parameters of s, v, a, and jerk respectively. sr is the reference driving distance, which is set to sr = 0.5 (start_s + end_s), that is, the expected s of the ego vehicle is the middle position of the lane change space; In the time range from t0 to tN, considering the constraints of maximum acceleration, maximum and minimum jerk, a quadratic optimization problem is constructed and solved to obtain the velocity curve of the ego vehicle; For each lane-changing space, the solver calculates the time t required for its speed curve to reach the safe lane-changing condition, and selects the lane-changing space with the smallest t as the optimal lane-changing space. Based on the speed curve corresponding to the optimal lane-changing space, the lane-changing timing and the acceleration and deceleration model of the vehicle to reach the appropriate timing are obtained, and the lane-changing strategy is generated to output the lane-changing trajectory.

10. The method for intelligent acceleration and deceleration lane changing decision-making for automatic driving according to claim 1, characterized in that: In actual driving scenarios, extreme scenarios are identified and processed. For example, when a dangerous lane change occurs due to sudden acceleration of the vehicle behind in the target lane or untimely upstream perception, the vehicle cancels the current lane change behavior. When changing lanes in a construction area, the vehicle triggers the lane change intention before approaching the construction area, and defines a lane risk model based on the obstacle status of the lane and the construction area. The lane risk model is combined at the lane change intention generation level to trigger the lane change to avoid the construction area.

Images